The two interrelated goals of paleoclimate research are to reconstruct past climatic conditions and to understand the dynamics of the global climate system. The enormous number of complex interactions operating on a wide variety of temporal and spatial scales within the climate system make this second goal one of the most monumental tasks in the physical sciences. However, this goal is also one of the most pressing, given the possibility that human activities may soon dramatically alter the planet's climate.
One of the most effective ways to study the dynamics of the climate system is to analyze time series of various climate variables. Time series provide information about the magnitude and rate of past climate change, climate responses to changes in incoming solar radiation (insolation), and connections between different parts of the climate system. The results presented in this dissertation advance paleoclimate research by providing several important tools for improving paleoclimate time series and by presenting a novel analysis of climate change over the last 5 million years (Myr).
The examples in this research primarily involve global climate change on timescales of 104 - 106 years, but many of the techniques developed in this research can quite easily by applied to other spatial or temporal scales. This work focuses on the Plio-Pleistocene because it is recent enough to provide information about modern climate dynamics and to have plenty of available data and because it is an interval in which the climate experiences several dramatic but poorly understood changes, specifically a long-term cooling trend and a switch in the period of glacial cycles.
To place this work in perspective, Section 1.1 establishes the importance of time series in paleoclimate studies. Section 1.2 examines the techniques for the creating paleoclimate age models, which are a common theme throughout this research. Next, Section 1.3 reviews unanswered questions about Plio-Pleistocene climate. Finally, the significant contributions from each chapter of this work are summarized in Section 1.4.
This research focuses on paleoclimate time series as an essential tool in the study of climate dynamics. While detailed descriptions of climate conditions at a single point in time are useful for providing ``snapshots'' of stable or transitional climate states, time series provide an opportunity to study the interactions of different parts of the climate systems. The number and complexity of these interactions are the main obstacles to fully modeling the climate system.
Time series have long played an important role in paleoclimate research. Even before continuous time series were available, geologists became aware of a series of glacial-interglacial cycles and proposed that they were forced by changes in the orbital configuration of the Earth that affected the seasonal and latitudinal distribution of insolation [Koppen and Wegener, 1924; Milankovitch, 1930]. Tremendous advancements in our understanding of past climates have occurred since the first recovery of long, continuous climate records from marine sediments several decades ago. For example, time series of global ice volume have confirmed that orbital changes play a key role in pacing glacial cycles [e.g., Hays et al., 1976; see Appendix A].
In the last two decades, the number of paleoclimate time series has virtually exploded. Time series relevant to paleoclimate study range in temporal resolution from weeks to millions of years, and measure such diverse variables as global ice volume, sea surface temperature (SST), floral and faunal distributions, wind strength, and atmospheric composition. Each new time series provides another clue to solving the climate puzzle, but the climate system is so vastly complex that many basic questions remain unanswered, such as those presented in Section 1.3 or how the climate will respond to anthropogenic emissions of carbon dioxide.
If we are ever to solve the climate puzzle, we must make the most of each puzzle piece that we find. This means seeing each piece clearly, determining how the pieces fit together, and viewing the results as a whole. In terms of time series, this means creating high-resolution, low-noise climate records, developing accurate age models which allow these records to be compared, and analyzing time series collectively. Creating age models for long paleoclimate records has always been a challenge, but the current profusion of age models (almost a different one for each new dataset) has created a major obstacle to the coordinated analysis of data from different regions.
My research addresses many steps in the creation and analysis of paleoclimate time series. Chapter 2 of this work presents a technique for placing diverse paleoclimate records on a common timescale. Chapter 3 presents a technique to improve the quality of time series derived from marine sediments and quantifies levels of distortion within these records. Chapter 4 aligns 57 globally-distributed marine climate records in order to produce a better estimate of global climate response, to constrain Plio-Pleistocene age models, and to allow for future analysis of spatial gradients in the ocean. Chapter 5 uses time series analysis of the global climate record from the previous chapter to describe changes in climate sensitivity during the Plio-Pleistocene. Finally, Chapter 6 relates the conclusions of this work to suggestions for future research.
One of the main challenges in the development and analysis of paleoclimate time series is age control. Precise age control is often important to determine the way in which two or more climate variables interact; however, very few techniques exist for directly determining the age of paleoclimate measurements. Most time series derive from the gradual accumulation of material, often sediment or ice, and different aspects of the climate are measured through the chemical or biological composition of the accumulated material. Unfortunately, direct dating techniques for most of these records have very low resolution before a few tens of kiloyears ago (ka), due to the limits of radiocarbon dating and layer counting in ice cores.
There are two main techniques for creating continuous age models when no high-resolution direct dating is available. One is to assume constant accumulation rates between coarsely spaced age control points, such as magnetic reversals. The other is to align or ``tune'' the climate record to match its assumed forcing, particularly changes in orbital configuration, as described in Appendix A. (Closely related to tuning is graphic correlation, in which an age model is derived from the assumption that one climate record closely resembles another.) Both age model techniques have major shortcomings, but encouragingly they often produce similar results [e.g., Imbrie et al., 1984; Huybers and Wunsch, 2004]. Because the development of age models and the variability of sedimentation rates are common themes throughout this work, a close examination of the assumptions and theories behind age model construction is presented below.
The validity of assuming constant sedimentation rates (CSR) between age control points depends on the depositional environment and the spacing of age control points. In many environments, such as river deltas or ice sheets, accumulation rates are known to be highly variable, so the CSR assumption is usually reserved for deep sea environments where sediment deposition is very slow and generally consists of wind-blown dust and the local production of carbonate and siliceous microfossils. In these environments the factors which could affect accumulation rates are wind direction and strength, dust supply due to continental aridity, biological productivity due to nutrient availability, and deep water chemistry which affects carbonate dissolution. Sedimentation rates will also be affected by the supply of ice-rafted detritus (IRD) at high-latitude sites and by deep water currents at drift sites. All of these factors are thought to exhibit some degree of variability, but the effect of this variability on marine sedimentation rates is very poorly constrained [e.g., Hagelberg et al., 1995a; this work, Chapter 3]. A better estimate of sedimenatation rate variability would be especially useful for graphic correlation (as in Chapter 2) and orbital tuning (as in Chapter 4).
Despite the fact that the CSR assumption is known to be somewhat inaccurate, it remains in use [e.g., Raymo and Nisancioglu, 2003; Ashkenazy and Tziperman, 2004] because the alternate technique of orbital tuning has its own significant sources of uncertainty. The two techniques used to minimize error in the CSR assumption are the selection of sites expected to exhibit the least variability (i.e., far from land and well above the lysocline) and the averaging of sedimentation rates from many different locations [e.g., Huybers and Wunsch, 2004; Lisiecki and Raymo, 2005]. The logic behind this second approach is that sedimentation rate changes at different sites are not strongly correlated so that variability will tend to cancel out in the averaging process.
Because this averaging process is applied in Chapter 4, it bears further scrutiny. The factors affecting sedimenation rates which would tend to be localized are wind and current patterns and productivity changes due to upwelling, SST, or thermocline depth. The accumulation of IRD tends to correlate across high-latitude sites, and carbonate dissolution rates correlate within ocean basins [e.g., Curry et al., 1988; Hagelberg et al., 1995b]. Changes in aridity and nutrient supply can probably be local or global. On long timescales, biological evolution will have both local and global effects. Given the number of spatially extensive factors, average sedimentation rates are unlikely to be absolutely constant, but the most stable averages would be derived from geographically diverse sites, especially including sites from both the Atlantic and Pacific and not including too many sites from high latitudes, where sedimentation rates are strongly affected by IRD [e.g., Karner et al., 2002].
To estimate the uncertainty of age models derived from sedimentation rates one must have estimates of the magnitude and timescale of variability in sedimentation rates. A formal analysis of this problem remains to be done, but the automated correlation technique presented in Chapter 2 may assist in such a study. Until such an analysis is performed, the uncertainty of CSR age models must be judged in relative terms. Globally averaged sedimentation rates should exhibit less variability than those from a single site, and the timescales of greatest variability in these averages are probably those associated with global climate change, such as the glacial-interglacial cycle.
Orbital tuning is the dating of paleoclimate records through correlation with proposed forcing functions, typically insolation or the orbital parameters which control its distribution (see Appendix A). The advantage of this dating technique is that it is very high resolution, but assuming a particular forcing function may introduce circular reasoning into the study of climate response. Other shortcomings of orbitally tuned age models are that they are susceptible to errors in the correlation of forcing and response and that they cannot provide information about the phase difference between forcing and response. The largest age model errors occur when a response is correlated to the wrong change in forcing (e.g., the wrong orbital cycle), but a potentially more pervasive error is overtuning, in which a climate signal is distorted in the process of maximizing its correlation to a forcing function. Because Chapters 4 and 5 rely on orbital tuning, each limitation of the technique is discussed below. The basic theory of orbital forcing is presented in Appendix A.
The circular resoning argument is the most fundamental challenge to the orbital tuning technique. However, excluding orbital tuning from all studies of orbital responses is unnecessary. Instead, conclusions based on orbitally tuned age models should be analyzed based on sensitivity to uncertainty in the data, as with any other scientific investigation. This presents the question of how the accuracy of orbitally tuned age models can be tested. The most convincing argument in favor of these age models is their close agreement with many CSR age models [e.g., Huybers and Wunsch, 2004], which are based on entirely independent assumptions. If the same result is obtained from both the CSR and tuned age models, one can be fairly confident of its conclusion, assuming that at least two independent age estimates are used in the construction of the CSR model. If a study's conclusion requires a tuned age model which produces some deviation from the CSR model, one must evaluate the probability that those deviations in sedimentation rate occurred. This would be one important application of a study of sedimentation rate variability.
Relationships between forcing and response that are not directly related to age model are also often used to validate orbitally tuned age models. For linear responses, amplitude modulation can be a good test of whether cycles in the forcing and response are correctly matched. The most frequently used test of orbital tuning is coherence with obliquity and precession; coherence measures the tendency of two functions to covary at specific frequency with some phase lag. However, significant coherence at multiple frequencies can be artificially introduced into a pure sine wave by overtuning [Shackleton et al., 1995a]. A better test of amplitude modulation can be calculated using complex demodulation [Shackleton et al., 1995a]. Alternately, overtuning can be identified by changes in sedimentation rate if expected changes can be constrained, for example by using globally averaged sedimentation rates. Changes in average sedimentation rate can also detect when a glacial cycle has been correlated to the wrong orbital cycle (e.g., Figure 4.10).
The lack of phase information provided by orbitally tuned age models is another limitation of the technique, but the associated uncertainties are fairly small, typically only a few kiloyears. Phases for orbital tuning are often selected based on physical arguments, such as estimates of system response times. However, the best phase estimates are derived from multiple climate proxies measured in the same core, which allows the relative phases of responses to be observed directly [e.g., Clemens et al., 1996]. Relative phases may also provide constraints on absolute phases because none of the observed responses should precede the forcing or lag it by more than a quarter cycle (assuming a direct response).
Graphic correlation is a technique for developing age models based on the correlation of one climate variable to another [e.g, Martinson et al., 1982; Prell et al., 1986]. In some cases, strong theoretical justifications exist for correlating the two time series, such as the measurement of atmospheric composition at two different locations. In other cases, correlations are proposed based on the observed similarity of two times series before a physical connection between the two is established (e.g., Figure 2.8). Both situations are fairly common and play important roles in advancing our understanding of the climate system by providing common age models and suggesting physical connections between different climate variables and geographic regions. Like orbital tuning, graphic correlation cannot provide information about the relative phases of two variables.
Because the number of paleoclimate time series is rapidly increasing, techniques for creating comparable age models should become ever more useful. The power of graphic correlation is that one well-dated record can provide age models for any others displaying the same pattern of variability, which is very common in the highly interconnected climate system. Even if no absolute age model is available, placing climate records on the same relative age model can be incredibly useful because it allows for measurements of geographic variability. In the case of marine sediment records, it can also help constrain age models based on the principal of reduced variability in average sedimentation rates (as in Chapter 4). Chapter 2 presents a new, automated graphic correlation technique, which allows researchers to quickly generate physically realistic correlations between paleoclimate time series.
There are three large, unanswered questions about Plio-Pleistocene climate. The first is what has driven the long-term global cooling trend of the last 4 Myr. The other two involve identification of the mechanisms responsible for generating 41-kyr and 100-kyr glacial cycles, respectively. While both periods occur in orbital configuration changes (see Appendix A), the dominant periodicity found in northern hemisphere summer insolation is 23 kyr, a cycle which is relatively weak in most high-latitude climate records.
Benthic d18O values increase during the Plio-Pleistocene [Zachos et al., 2001; this work, Chapter 5], likely indicating both a decrease in ocean temperatures and and increase in global ice volume (see Appendix B). The reasons for this climatic cooling remain unclear. While a decline in atmospheric concentrations of greenhouse gases could produce such a cooling, research suggests that CO2 levels have remained relaively stable since 20 Ma [Pearson and Palmer, 2000]. Another theory links tectonic changes in ocean gateways to global climate change. For example, closing of the Panamanian gateway at ~4 Ma may have had a dramatic effect on ocean circulation patterns, increasing the amount of moisture available for producing northern ice sheets [Haug and Tiedemann, 1998]. A delay between the gateway closing and large ice sheet formation may be related to modulations in the amplitude of obliquity, but this theory still provides no mechanism to explain continued cooling after 2.6 Ma. For a more detailed description of long-term trends in Plio-Pleistocene climate, see Chapter 5.
Virtually all models of the 100-kyr glacial cycle in the late Pleistocene rely on the argument that long time constants in the climate system favor a 100-kyr glacial cycle by decreasing the climate's sensitivity to the more rapid variations of precession and obliquity. Theories differ in the source of the long time constant, but popular suggestions are large ice sheets and their underlying bedrock [Weertman, 1964] or the global carbon cycle [Shackleton, 2000]. A direct response to eccentricity is unlikely because it has very little net effect on annual insolation and because the 400-kyr power of eccentricity produces little response in late Pleistocene climate.
Most models differ in the extent to which eccentricity and nonlinear responses to obliquity and precession drive 100-kyr glacial cycles. One group of theories holds that 100-kyr oscillations are self-sustained, requiring no external forcing [e.g., Maasch and Saltzman, 1990; Tziperman and Gildor, 2003]. Even if variability is self-sustained, orbital forcing may still play an important role in pacing 100-kyr glacial cycles [Huybers and Wunsch, 2005]. Another group of theories posits that nonlinear responses to obliquity and/or precession result in conditions conducive to deglaciation with a period of approximately 100 kyr [e.g., Raymo, 1997; Paillard, 1998; Ruddiman, 2003]. Finally, it is possible to produce 100-kyr power in climate models simply by introducing different rates of ice sheet growth and decay [Imbrie and Imbrie, 1980]. One very different model proposes that 100-kyr cycles in orbital inclination may force late Pleistocene glacial cycles [Muller and MacDonald, 1997], but very little support exists for forcing mechanisms associated with inclination [Winckler et al., 2004].
Another major puzzle is why 100-kyr power increased so dramatically at 0.8 Ma, despite no apparent change in orbital forcing. (However, Chapter 5 will propose this change may result from a decrease in the 100-kyr power of eccentricity at this time.) Most theories attempt to relate the appearance of 100-kyr power with a coeval increase in ice sheet size [e.g., Paillard, 1998; Ashkenazy and Tziperman, 2004]. This increase in ice sheet size at 0.8 Ma might simply result from the long-term trend of global cooling, but Clark and Pollard [1998] propose a specific mechanism to increase ice volume through the gradual erosion of continental regolith during repeated glaciations. The results presented in Chapter 5 do not favor the appearance of large ice sheets as a causal mechanism for the switch to 100-kyr cycles at 0.8 Ma because 100-kyr power is also observed in the climate record at 3 Ma.
Strong obliquity responses in the late Pliocene and early Pleistocene may be associated with the seasonal and spatial distributions of insolation specific to obliquity cycles. Raymo and Nisancioglu [2003] propose that the insolation gradients between high and low latitudes, which are closely tied to orbital obliquity, may control continental ice volume by altering poleward moisture transport. Philander and Fedorov [2003] propose that sensitivity to obliquity results from the thermocline's response time of several decades. Because obliquity has a small net effect on annual insolation as a function of latitude which precession lacks, small annual imbalances in the regional heat budget of the thermocline could accumulate to produce 41-kyr cycles in SST and poleward ocean heat transport.
Finally, 41-kyr glacial cycles could result from the same kind of self-sustained variability (SSV) proposed to produce 100-kyr cycles [Ashkenazy and Tziperman, 2004]. This theory proposes that obliquity merely paces glacial cycles, perhaps through the formation of sea ice. Such a mechanism could produce an abrupt switch from 41-kyr to ~100-kyr glacial cycles simply by crossing a threshold in maximum ice volume. The results presented in Chapter 5 demonstrate that SSV is a plausible source of 41-kyr power only after 1.4 Ma.
Graphic correlation is a powerful tool which is often applied to paleoclimate records and is especially well-suited to orbital- and millennial-scale climate changes. Given the increasing number of paleoclimate records available today, the need for fast accurate correlation is especially urgent. However, previously available correlation techniques had serious shortcomings. The oldest technique of manual correlation is extremely time consuming and often has no formalized constraints. The other commonly used technique is an automated correlation algorithm [Martinson et al., 1982] based on using sine waves to construct a mapping function which optimizes the correlation between time series. The two shortcomings of this technique are poor constraints on sedimentation rates, which may lead to overtuning, and an inability to find the globally optimal solution, requiring the age models of the two time series to be fairly similar before performing the automated correlation.
Chapter 2 of this research presents a new technique for performing graphic correlations, which is the first automated algorithm designed for paleoclimate time series that constrains sedimentation rates and finds a globally optimal solution. As discussed extensively above, sedimentation rates play an important role in assessing the accuracy of an age model and preventing overtuning that can produce misleading results. The automation and global solution of this algorithm are generally issues of convenience which will hopefully allow more paleoclimate time series to be placed on a common timescale by making the process faster and easier. Two practical applications of this algorithm are presented in Chapters 3 and 4.
In Chapter 3 the graphic correlation algorithm is adapted to automate the generation composite depth sections for marine sediments cores and to characterize patterns of core deformation which can distort paleoceanographic records. The new compositing algorithm addresses two shortcomings of the standard splicing technique [Hagelberg et al., 1992]: (1) a failure to correct for deformation within cores, so that one sedimentary feature may have a slightly different composite depth in each hole, and (2) the tendency to produce composite depths which are ~10% greater than recorded drill depths. This chapter also examines 618 cores from Ocean Drilling Program (ODP) Legs 138 and 154 to present the first estimates of systematic core deformation in ODP cores. Finally, we analyze the factors responsible for producing depth inflation in traditional composite depth scales.
The construction of d18O stacks is a traditional technique for improving the signal quality of the d18O record, which describes changes in global ice volume and ocean temperature (see Appendix B). Chapter 4 presents a new, 5.3-Myr stack of 57 aligned benthic d18O records [the ``LR04'' stack; Lisiecki and Raymo, 2005] and an improved d18O age model. This research provides the paleoclimate community with important stratigraphic tools to describe Plio-Pleistocene climate change and to aid in the comparison of widely distributed marine climate records. The LR04 stack is the first composed of more than three benthic records to extend beyond 850 ka, and its improved signal quality is used to identify 24 new marine isotope stages in the early Pliocene. An improved Plio-Pleistocene age model is created by tuning the stack to a simple ice model based on June 21 insolation at 65oN while minimizing changes globally averaged sedimentation rates. Finally, the relative phases of obliquity and precession response are analyzed to constrain their response times and their sensitivity to northern hemisphere forcing.
Climate transitions are a popular topic in paleoclimate research because identifying the causes of climate change provides a great deal of insight into the dynamics of the climate system. The Plio-Pleistocene is a time of many climate transitions, including a gradual cooling trend, the onset of NHG, and the transition from 41-kyr to 100-kyr glacial cycles. Each of these transitions has been intensively studied, but none are well understood.
The analysis presented in Chapter 5 differs from previous studies in several key ways. First, climate change is described by the 5.3-Myr benthic stack developed in Chapter 4; the use of this stack improves results by increasing the signal-to-noise ratio of the climate signal and by providing a robust age model. Second, the amplitude modulation and cycle shape of climate response are measured as functions of time through the entire Plio-Pleistocene, which aids in the recognition of trends and correlations. Finally, the critical innovation in this study is the description of a long-term ``amplification'' of climate sensitivity, as if the climate system were a stereo whose volume was slowly and steadily increased. When an exponential amplification is removed from the climate record, the relationship between orbital forcing and climate response is strikingly clear.
Many studies have noted that the power of obliquity response has been relatively steady over the last 1 or 2 Myr [e.g., Imbrie et al., 1993; Liu and Herbert, 2004], and some have even remarked that this is unusual in light of changes in the amplitude of obliquity forcing [e.g., Ruddiman et al., 1989]. However, this is the first study to demonstrate a strongly linear sensitivity to obliquity forcing from 5 - 1.4 Ma by removing an exponential trend in the power of obliquity response. The significance of this finding is two-fold. (1) It constrains the possible mechanisms involved in generating the 41-kyr glacial cycle, actually suggesting a change in mechanism when response switches from linear to nonlinear. (2) It identifies a heretofore unrecognized climate transition at 1.4 Ma. This transition impacts both the amplitude of obliquity response and the shape of glacial cycles, as evidenced by a dramatic decrease in the relative duration of interglacial stages. The combined onset of nonlinear obliquity response and less stable interglacials suggests a major change in the physics of the climate system, but the mechanisms responsible for this transition have yet to be identified.
Many other new and significant results are presented in Chapter 5. (1) The linearity of precession response is verified for the entire Plio-Pleistocene, but apparently it does not experience exponential amplification until 2.5 Ma. (2) The first appearance of saw-tooth asymmetry in glacial cycles is identified at 2.5 Ma, shortly after the onset of NHG. (3) 100-kyr power appears in the benthic d18O record by at least 3 Ma and is anticorrelated with the 100-kyr power of eccentricity. These results suggest that the 100-kyr glacial cycle is impeded by rather than forced by the 100-kyr modulation of precession and that 100-kyr glacial cycles are not dependent upon the physics of large northern hemisphere ice sheets. (4) The mid-Pleistocene transition is not associated with any major changes in glacial shape or precession and obliquity response. The climate's rapid increase in 100-kyr power at 0.8 Ma may result from a negative correlation with precession modulation superimposed on a 5-Myr trend of increasing 100-kyr power.
Copyright 2005 by Lorraine E. Lisiecki. Do not reproduce without permission.